category: model check
base models: QWEN-Image MMDiT by Alibaba Cloud’s Tongyi Qianwen
workflow: powered by ComfyUI
influenced: by QWEN-Image-Lightning LoRA
model released: 2025-08
.
All generations using ComfyUI.
QWEN MMDiT by Alibaba’s Tongyi Qianwen with the following downloads: qwen_image_fp8_e4m3fn.safetensors > diffusion_model
qwen_2.5_vl_7b_fp8_scaled.safetensors > text_encoder
qwen_image_vae.safetensors > vae
Qwen-Image-Lightning-8steps-V2.0.safetensors > LoRA
“Qwen-Image has 20 billion parameters and uses a MMDiT (Multimodal Diffusion Transformer) architecture. The design goals are
(1) Complex, multilangual text rendering, and
(2) Strong alignment between the prompts and the generated images.”
Find this the workflow JSON and more information at: STABLEDIFFUISIONART.com
https://stable-diffusion-art.com/qwen-image/
Additional information at AIBase.com
https://www.aibase.com/news/20217
QWEN is open-source under the Apache 2.0 license, free for commercial use.
Hugging Face model card: https://huggingface.co/Qwen/Qwen-Image
Specifications: LoRA QWEN-Image-Lightning strenght 1.00 – steps 13 – cfg 1.6 – euler / simple – denoise 0.99. Format is 1088×1088 px.
Slight color correction and eye optimisation:
Original:
The result is somewhat soft, but I like it. Anyway, the next steps are about adding more contrast and sharpness to the picture. The hands and fingers are well-shaped!
Testing different steps:
1) #LoRA QWEN-Image-Lightning strength 1.00 – shift 4.5 – steps 9 – cfg 1.6 – euler / simple – denoise 0.99
2) #LoRA QWEN-Image-Lightning strength 1.00 – shift 4.5 – steps 18 – cfg 1.6 – euler / simple – denoise 0.99
Testing different steps, without LoRA:
1) no LoRA – shift 3.1 – steps 20 – cfg 1.0 – euler / simple – denoise 0.99
2) no LoRA – shift 3.1 – steps 30 – cfg 1.0 – euler / simple – denoise 0.99
3) no LoRA – shift 3.1 – steps 40 – cfg 2.5 – euler / simple – denoise 0.99 (team’s suggestion)
